Friday 28 June 2019

Choosing a new VLE: Technological Uncertainty points towards Personal Machine Learning

I sat in a presentation from a company wanting to replace my university's Virtual Learning Environment yesterday. It was a slick presentation (well practiced) and people generally liked it because the software wasn't too clunky. Lack of clunkiness is a sign of quality these days. New educational technology's functionality is presented as a solution to problems which have been created by other technology, whether it is the management of video, the coordination of marks, mobile apps, management of threaded discussion, integration of external tools, and so on. A solution to these problems creates new options for doing the same kinds of things: "use our video service to make the management of video easier", "use our PDF annotation tool to integrate with our analytics tools", etc. Redundancy of functionality is increased in the name of simplification of technological complexity in the institution. But in the end, it can't keep up: what we end up with is another option to choose from, an increase in the uncertainty of learners and teachers, which inevitably necessitates a managerial diktat to insist on the use of tool x rather than tool y. Technology that promises freedom produces restriction, and an increasingly wide stand-off between the technology of the institution and the technology of the outside world.

The basic thesis of my book "Uncertain Education" is that technology always creates new options for humans to act. Through this, the basic human problem of choosing the manner and means of acting becomes less certain. Institutions react to rising uncertainty in their environment often by co-opting technologies to reinforce their existing structures: so "institutional" tools, rather than personal tools, dominate. Hence we have corporate learning platforms in universities, and the dominance of corporate online platforms everywhere else. This is shown in the diagram below: the "institution's assistance" operates at a higher-level "metasystem", which tries to attenuate the uncertainty of learners and teachers in the primary system (the circle in the middle). Institutional technology like this seeks to ease the burden of choice of technology for workers, but the co-opting institutional process can't keep up with the pace of change in the outside world - indeed, it feeds that change. This situation is inherently unstable, and will, I think, eventually lead to transformation of organisational structures. New kinds of tools may drive this process. I am wondering whether personal AI, or more specifically, personal machine learning, might provide a key to transformation.

Machine learning appears to be a tool which also generates many new options for acting. Therefore it also should exacerbate uncertainty. But is there a point at which the tools which generate new options for acting create new ways in which options might be chosen by an individual? Is there a point at which this kind of technology is able assist in the creation of a coherent understanding of the world in the face of explosive complexification produced by technology? One of the ways this might work is if machine learning tools could assist in stimulating and coordinating conversations directly between teachers and learners. Rather than an institutional metasystem, machine learning could operate at the level of the human system in the middle, helping to mitigate the uncertainty that is required to be managed by the higher level system:

Without wanting to sound too posthuman, machine learning may not be so much a tool as an evolutionary "moment" in the relationship between humans and machines. It is the moment when the separation between humans and machines, which humans have defended since the industrial revolution in what philosopher Gilbert Simondon calls "facile humanism", becomes indefensible. Perhaps people like Friedrich Kittler and Erich Horl are right: we are no longer humans selves constituted of cells and psychology existing in a techno-social system; now the technical system constitutes the human "I" in a process intermediated by our cells and our consciousness.

I wonder if the point is driven home when we appreciate machine learning tools as an anticipatory system. Biological life drives an anticipatory process in modelling and adapting to the environment. We don't know how anticipation occurs, but we do possess models of what it might be like. One way of thinking about anticipation is to imagine it as a kind of fractal - something which approximates to David Bohm's 'implicate order' - an underlying and repeated symmetry. We see it in nature, in trees, in art, music, and in biological developmental processes. Biological processes also appear to be endosymbiotic - they absorb elements of the environment within their internal structure, repeating them at higher levels of organisation. So cells absorbed the mitochondria which once lived independently, and the whole reproduces itself at a higher order. This is a fractal.

Nobody quite knows how machine learning works. But the suspicion is that it too is a fractal. Machine learning anticipates the properties of an object it is presented with by mapping its features which are detected through progressive layers of analysis focusing on smaller and smaller chunks of an image. The fractal is created by recursively exploring the relationship between images and labels across different levels of analysis. Human judgements which feed the "training" of this system eventually become encoded as a set of "fixed points" in a relational pattern in the machines model.

I don't think we've yet grasped what this means. At the moment we see machine learning as another "tool". The problem with machine learning as a "tool" is that it is then used to provide an "answer": that is, it is used to filter-out information which does not relate to this answer. Most of our "information tools", which provide us with increased options for doing things, actually operate like this: they discard information, removing context. This adds to the uncertainty they produce: tool x and tool y both do similar jobs, but they filter out different information. Choosing which tool to use is to decide which information we don't need, which requires human anticipation of an unknowable future. Fundamentally, this is the problem that any university wanting to invest in new technology is faced with. Context is everything, and identifying the context requires anticipation.

Humans are "black boxes": we don't really know how any of us work. But as black boxes who converse, we gradually tune-in to each other, understanding the behaviour of each of us, and in the process, understanding more about our own "black box". In the process we manage the uncertainty of our own existence. Machine learning is also a black box. So might the same thing work? If you put two black boxes together, do they begin to "understand" each other? If you put a human black box together with a machine black box, does the human gain insight into the machine, and insight into the themselves through exploring the operation of the anticipatory system in the machine? If you put a number of human black boxes together with a machine black box, does it stimulate conversation between the humans as well as engagement with the machine? It is important to note in each of these scenarios, information is preserved: context is maintained with the increase in insight, and can be further encoded by the machine to enrich human conversation.

I wonder if these questions point to a new kind of organisational setup in institutions between humans and technology. I cannot see how the institutional platform can really be a viable option for the future: discarding information is not a way forward. But we need to understand the nature of machine learning, and the ways in which information can be preserved in the human machine relationship.

Tuesday 18 June 2019

Machine Learning as a Personal Anticipatory System

Can a living system survive without anticipation? As humans we take anticipation for granted as a function of consciousness: without an ability to make sense of the world around us, and to preempt changes, we would not be able to survive. We attribute this ability to high-level functions like language and communication. At the same time, the ability of all living things to adapt to environments whilst not always showing the same skill of language is apparent, although many scientists are reluctant to attribute consciousness to bacteria or cells. Ironically, this reluctance probably has more to do with our human language for describing consciousness, than it does to the nature of any "language" or "communication" of cells or bacteria!

We believe human consciousness is special, or exceptional, partly because we have developed a language for making distinctions about consciousness which reinforces a separation between human thought and other features of the natural world. In philosophy, the distinction boils down to "mind" and "body". We have now reached a stage of development where continuing to think like this will most likely destroy our environment, and us with it.

Human technology is a product of human thought. We might believe our computers and big data to be somehow "objective" and separate from us, but we are looking at the manifestations of consciousness. Like other manifestations of consciousness such as art, music, mathematics and science, our technologies tell us something about how consciousness works: they carry an imprint of consciousness in their structure. This is perhaps easiest to see in the artifice of mathematics, which whilst being an abstraction, appears to reveal fundamental patterns which are reproduced throughout nature. Fractals, and the imaginary numbers upon which they sit, are good examples of this.

It is also apparent in our technologies of machine learning. Behind the excitement about AI and machine learning lies a fundamental problem of perception: these tools display remarkable properties in their ability to record patterns of human judgement and reproduce them, but we have little understanding of how it works. Of course, we can describe the architecture of a convolutional neural network (for example), but in terms of what is encoded in the network, how it is encoded, and how results are produced, we have little understanding. Work with these algorithms is predominantly empirical, not theoretical. Computer programmers have developed "tricks" for training networks, such as training a full network with existing public domain image sets (using, for example, the VGG16 model), but then retraining the bottom layer for the specific images that they want identified (for example, images of diabetic retinopathy, or faces). This works better than training the whole network on specific images. Why? We don't know - it just does.

It seems likely that whatever is happening in a neural network is some kind of fractal. The training process of back-propagation involves recursive processing which seeks fixed points in the production of results across a vast range of variables from one layer of the network to the next. The fractal nature of the network means that retraining the network cannot be achieved by tweaking a single variable: the whole network must be retrained. Neural networks are very dissimilar from human brains in this way. But the fractal nature of neural networks does raise a question as to whether the structure of human consciousness is also fractal.

There is an important reason for thinking that it might be. Fractals are by definition self-similar, and self-similarity means that a pattern perceived at one level with one set of variables can be reproduced at another level, with a different set of variables. In other words, a fractal representation of one set of events can have the same structure as the fractal pattern of a different set of events: perception of the first set can anticipate the second set.

I've been fascinated by the work of Daniel Dubois on Anticipatory Systems recently partly because it is closely related to fractals, and it also seems to have a strong correlation to the way that neural networks work. Dubois makes the point that an anticipatory system processes events over time by developing models that anticipate them, whilst also generate multiple possible models and selecting the best fit. Each of these models is a differently-generated fractal.

If we want to understand what AI and machine learning really mean for society, we need to think about what use an artificial anticipatory system might be. One dystopian view is that it means the "Minority Report" - total anticipatory surveillance. I am sceptical about this, because an artificial anticipatory system is not a human system: its fractals are rigid and inflexible. Human anticipation and machine anticipation need to work together. But a personal artificial anticipatory system is something that is much more interesting. This is a system which processes the immediate information flows of experience and detects patterns. Could such a system help individuals establish deeper coherence in their understanding and action? It might. Indeed, it might counter the deep dislocation produced by overwhelming information that we are currently immersed in, and provide a context for a deeper conversation about understanding.


Sunday 16 June 2019

Machine Learning and the Future of Work: Why eventually we will all create our own AIs

I'm on my way to Russia again. I've had an amazing couple of days with a Chinese delegation from Xiamen Eye Hospital and the leading experts in retinal disease in China, who are collaborating with us on a big EPSRC project. There was a very special atmosphere: despite the language differences, we were all conscious of staring at the future of medical diagnostics where AI and humans work in partnership.

There's a lot of critical dystopian stuff about technology in society and education in the e-learning discourse at the moment. I think history will see this critical reaction more as a response to desperately nasty things going on in our universities, rather than an accurate prediction of the future. I am also subject to these institutional pathologies, but I suspect both the dystopian critiques and the institutional self-harm are symptoms of more profound changes which are going to hit us. Eventually we will rediscover a sane way of organising human thought and creativity once more, which is what our universities used to do for society.

So this is what I'm going to say the students in Vladivostok:

Machine Learning, Scientific Dialogue and the Future of Work
It is not unusual today to hear people say how the next wave of the technological revolution will be Artificial Intelligence. Sometimes this is called the "4th industrial revolution": there will be robots everywhere - robot teachers, robot doctors, robot lawyers, etc. In this imagined future, machines are envisaged to take the place of humans. But this is misleading. The future will however involve a deeper partnership between humans and intelligent machines. In order to understand this, it is important to understand how our technologies of AI work, how the processes of creating AIs and machine learning are becoming available to you and me, and how human work is likely to change in the face of technologies which have remarkable new capabilities. 
In this presentation, I will explain how it will become increasingly easy to create our own AIs. Even now, the technologies of Machine Learning are widely available, increasingly standardised and accessible to people with a bit of computer programming knowledge. The situation at the moment is very much like the early web in the 1990s, when to create a website, people needed a bit of knowledge of HTML. As with the web, creating our own AIs will become something everyone can do.  
Drawing on my work, I will explain how in a world of networked services, there is one feature about Artificial Intelligence which is largely ignored by those not informed of its technical nature: AI does not need to be centralised. A machine learning algorithm is essentially a single (and often not very large) file, which can be embedded in any individual device (this is how, for example, the facial recognition works on your phone). The world of AI will be increasingly distributed. 
Finally, I will consider what this future means for human work. One of the important distinctions between human decision-making and AI is that humans make judgements in a context; AI, however, ignores context. In other words, AI, like much information technology, actually discards information, and this has many negative consequences on the organisation of institutions, stable society and the economy. The most potentially powerful feature of AI in partnership with humans is that it can preserve information by preserving the context of human judgement. I will discuss ways in which this can be done, and why it means that those things which humans do best – empathy, criticality, creativity and conversation – will become the essence of the work we do in the future.

Tuesday 4 June 2019

German Media Theory and Education

I'm discovering a branch of media studies which I was unaware of before Steve Watson pointed me to Erich Hörl's "Sacred Channels: The Archaic illusion of Communication". Hörl's book is amazing: cybernetics, Luhmann, Bataille, Simondon & co all spiralling around a principal thesis that communication is an illusion, and that many of our current problems arise from the fact that we don't think it is. The "illusion" of communication is very similar to David Bohm's assertion that "everything is produced by thought, but thought says it didn't do it". This is not "media studies" as we know it in UK universities. But it is how the Germans do it, and have been doing it for some time.

Just as Luhmann has been a staple of the German sociology landscape for undergraduate sociologists for 20 years now, so Luhmann's thinking informed a radical view of media which Hörl has inherited. He got it from Friedrich Kittler. Kittler died in 2011, leaving behind a body of work which teased apart the boundaries between media and human being. Most importantly, he overturned the hypothesis of Marshall McLuhan that media "extend" the human. Echoing Luhmann, Kittler says that media make humans. Just as Luhmann pokes the distinction between psychology and sociology (he really doesn't believe in psychology), Kittler dissolves the "interface" between the human and the media.

The result is that practically everything counts as media. Wagner's Bayreuth was media (Kittler wrote extensively about music, culminating with a four volume work he never finished, "Music and Mathematics"), AI is media, the city is media. So is education media? Not just the media that education uses to teach (which educational technologists know all about). But education itself - the systemic enveloping of conversations between students and teachers - is that media?

As Erich Hörl has pointed out, these ideas are very similar to those of another voice in technology studies who is gaining an increasingly dominant following after his death, Gilbert Simondon. Like Kittler, Simondon starts with systems and cybernetics. Simondon's relevance to the question of education and technology is quite fundamental. Kittler, I don't think, knew his work well, and Hörl acknowledges that he has further to go in his own absorption of the work. Simondon made a fundamental connection between media, or machine, and human beings as distinction-making, individuating entities. The individuation process - that process which Jung saw as the fundamental process of personal growth - was tied-up with the process of accommodating ourselves to the media which comprise us. This accommodation was achieved through levels of "transduction" - the multiple processes which produce multiple levels of distinctions, from the distinctions between our cells, to the distinctions in our language, and the distinctions with our environment. What happens in education, basically, is that the media which make us us are transformed through changes in the ways the transductions are organised at different levels.

I described a lot of this in my book, albeit not in the elegant fashion that Kittler, Hörl  (or Simondon) would have done. Kittler, Simondon and Hörl have got me thinking in a new way about how we think about education. There's much more to say about this however, because Kittler and Hörl's approach opens the way for a more empirical approach to understanding education as media. I was privileged to have learnt about Luhmann through one of his best disciples, Loet Leydesdorff. Leydesdorff's work has been dedicated to making Luhmann's theory empirically useful, which he has done by relating it to Shannon (which Luhmann did in the first place), and to the mathematics of anticipation by the Belgian mathematician, Daniel Dubois.

Here, we may yet have a science of education which straddles the boundaries between technology, critique, pedagogy and phenomenology whilst maintaining an empirical focus and theoretical coherence. That is the best way of getting better education. This science of education may well turn out to be exactly the same as the empirical and coherent science of media that Kittler and Hörl are aiming for, which transcends the sociological critique of media (seeing that as simply more media!), by providing a meta-methodology for making meaningful distinctions about our distinction-making processes in our media-immersed state.

Sunday 2 June 2019

Two kinds of information in music and media

My recent music has been exploring the idea that there are two kinds of information in the world. I am following the theory of my colleague Peter Rowlands, who had this to say (in the video below) on the subject of how nature is a kind of information system, but very different from the information systems of our digital computers. Peter summarises the difference by saying that digital information is made from 1s and 0s, but the significant thing is the 1. Nature, he contends, operates with multiple levels of zero. His reasons for thinking this are a thoroughly worked-through mathematical account of quantum mechanics, and particularly the Dirac equation (the only equation in Westminster Abbey!). Nature is all "Much ado about nothing".


I've been fascinated by "nothing" for a long time. Nothing is "absence" as opposed to "presence", and absence is (according to philosophers like Roy Bhaskar, cyberneticians like Gregory Bateson, and biologists like Terry Deacon) constraint. Constraint is important in digital information because it is represented by Shannon's concept of "redundancy". So there is a connection between nothing and redundancy. This resonates with me with something like music, because it is so full of redundancy, and music does appear to be "much ado about nothing".

There is something we do when we make music which somehow makes sense. The patterns we create create the conditions for richer patterns which eventually define a structure. Musicians create redundancy in the form of repetition which brings coherence to the music. There are different kinds of redundancy: pitch, rhythm, timbre, intervals, etc. Much of this patterning occurs in the context of an external nature which is always shifting the context in which the music is made. It might be the sound of the wind, or water, or traffic, computer sounds, or elevator music - our sonic environment is moving around us all the time. The musical sense may be the natural pattern-making response to this which seeks to produce coherence. If this is the case, then birdsong and the noises of all animals, and maybe even language itself, can be seen as a process of maintaining coherence of perception within an environment. This is a radical view when applied to language - it means that we don't communicate. We don't transfer "information" between us. As Niklas Luhmann says in his most famous quote,
"Humans cannot communicate; not even their brains can communicate; not even their conscious minds can communicate. Only communication can communicate."
He could be right. It's also quoted in Erich Hörl's new book "Sacred Channels: The archaic illusion of communication". Hörl follows a line of inquiry from Friedrich Kittler (who is also new to me) who argued that "media studies" needs to reject Marshall McLuhan's view that media extend the human; media makes the human. Gilbert Simondon said the same thing in connecting technology with human individuation. If there is a new theoretical way forwards for our thinking about technology, media and education, it rests with these people. Cybernetics is at the heart of it.

My music works with this idea where the electronic component of this piece represents the unstable shifting lifeworld of nature. Because this is "digital", we might think about it being not only the noise of the wind, but the noise of computers - digital information. The piano represents the musician's attempt to create pattern and maintain coherence in the whole. It is engaged in much ado about nothing.

Saturday 1 June 2019

Augar's Intergenerational Conversation

"Education" as a topic is very complex and hard to define. We might think of schools, classrooms, teachers, but whatever we choose to include as "education" inevitably excludes something. This is the problem of making a distinction about anything - but it is exacerbated when we think of education. The exclusion/inclusion problem creates uncertainty, and this uncertainty has to be managed through a process which usually involves talking to each other. Since talking to each other about important things is something we do in education, the topic of "education" is uniquely caught in a web of conversation. At the beginning of my book, I quoted Everett Hughes, who I think gets it about right when he says that education is a "complex of arts" where:
"the manner of practicing them is the very stuff of the clash of wills and interests; thus, the stuff of politics."
This is the same confrontation of wills and interests that parents face with their children, that the younger generation faces with the older. But all the way through, it is conversation which is the process of negotiation.

Philip Augar's review of post-18 education funding has been fairly warmly received - partly because of the thoughtful tone it sets, and its modest reprimands against some of the more outrageous excesses of marketised higher education. However, as many commentators have pointed out, in cutting the headline fee for students, but increasing the repayment period, it is appears more socially regressive than the current system. The message hasn't changed: it is the job of job of students (the young) to pay for their education (pay for their elders to teach them) over the course of their lives, although it is recommended that the loan funding to pay for education may be available for more flexible study options. The rationale is that the young benefit from education financially.

This week I've been involved in two separate discussions about the future of work. That Artificial Intelligence and global data is going to transform the workplace is barely beyond doubt. Exactly what kind of impact it will have on opportunities for the young is as yet unclear. Will every automated service create an equivalent number of jobs in other areas? Will the growth of profits of large corporations which benefit from a falling salary bill trickle-down to those left behind in the rush to reduce expensive human labour? Or are we heading for a data-coordinated future of globalised gig-work at globalised rock-bottom wages? If this is the future for the young, who could blame them for questioning the fairness of the financial burden they bear for an education which turns out to fall short of the promises made by their universities?

This is how we depress the future. As Stafford Beer said (in an unpublished notebook):
"In a hundred years from any `now', everyone alive will be dead: it would therefore be possible for the human race to run its affairs quite differently - in a wise and benevolent fashion. Education exists to make sure this does not happen."
 What is AI? What is the Web? Are they technologies "for organising our affairs quite differently"? They could be. "In a wise and benevolent fashion"? Not currently, according to Tim Berners-Lee and many others, but they could be. Then we come to education. Beer is making a point about education's role in reproducing social class divisions, which Bourdieu famously explained. But education is conversation, and more importantly, an intergenerational conversation. Our technologies are tools which both afford the coordination of conversation, and create new kind of remarkable artefacts for us to talk about. And these conversations are intergenerational: to be able to summon-up movies, videos or documents on demand and watch/read them together, whether online or together in the living room with our kids is profound and powerful. Something very special happens in those conversations.

In these kinds of simple things - of the elders sharing resources and talking with the young - there is something very important that we've missed in our educational market. Teaching involves the revealing of one's understanding, and the existential need to teach may lie with the elders, not the young. The gains for the young to participate are not always obvious to them (or anyone else). Promises made by the elders to the young about future riches are not always believable, but behind them lies the desire of the elders to encourage the young and preserve humanity after the elders are dead. Successful companies understand the importance of supporting the next generation, and they don't do it for the future financial benefit of the young. They do it to preserve the viability of the business.

If the existential need is to teach, not for the young to learn for future financial gain, then the elders should pay the young to be taught, for them to reveal their understanding to the next generation before the elders die. Only seeing it this way round makes any sense looking into the future: the young will have their own children, they will become the elders, they will have an existential need to teach, and they will pay their young to learn. The spirit of encouragement drives one generation to the next.

Now look at what Augar has tweaked but otherwise left untouched. Despite some florid prose extolling the virtues of education, the underlying existential issue is financial gain for the young through the acquisition of knowledge and certificates. The elders (of whom Augar is one) are merely functionaries in delivering knowledge and certificates. The promise of financial gain will be broken amidst employment insecurity, rents, lifelong debt and inequality. They will look at the elders and see their big houses and long lifespans (damn it, they won't even die quickly and leave an inheritance!), and ask how it is that their hopes for the future were diminished. Their only respite will be to inflict a similar injustice on their own children as they mutter "there is no alternative". This is positive feedback: the spirit of despair infects one generation to the next.

Augar's report is thoughtful though, so I don't want to dismiss it. One of his targets is the breaking down of the monolith of the 3-year degree course, and reconfiguring the way the institution's transactions with its students work. This is good. But Corbyn was right about the financing of education and who should pay. It's not just an argument about one generation of students. It's an argument about a viable society.